A major objective of real-time systems is to meet the timing requirements. A distributed transaction processing ensures a correct completion, which means that either a program is executed successfully, producing the intended results, or it has no results at all. If real-time applications run on distributed transaction processing environment, both correct completion and timeliness should be satisfied. Most previous commit protocols, however, fail to satisfy the timeliness for real-time processing. That is, methods for the timely completion do not always guarantee the correct completion or vice versa.
In this thesis, we propose new strategies that enforces both the timely and correct completion in the distributed real-time database systems. The strategies are composed of three parts: the language construct to specify the real-time requirements, the scheduling policy to support the timely completion, and the commit procedures to satisfy the atomicity.
The language constructs have three features: the specification rules of time constraints, the asynchronous communication primitives, and the registration method of the exception handling routines. Based on these language constructs, we propose a scheduling method to ensure the timeliness of real-time transactions and a commit protocol to guarantee the correct completion among the distributed transactions.
The scheduling method consists of two components: the one is real-time transaction scheduling that is related to task scheduling in real-time environment and the other is concurrency control that can be considered as operation level scheduling. To increase the success ratio of the urgent transactions, we propose the multi-level priority scheduling (MLPS) method. The basic idea, here, is the use of the criticalness of transactions with time information. We compare several scheduling policies with the MLPS. To enforce the data consistency, we propose a concurrency control method, called the Commit Timestamp Ordering P...